信用卡債務協商違約風險模型建構與運用--以國內某發卡銀行為例

Credit Card Debt Relief Program Default Risk Model Building and Application--The Case of Domestic Credit Card Issuer

林志強(Chih-Chian Lin)

指導教授 ： 梁世安

國立臺北大學/商學院/國際財務金融碩士在職專班/碩士(2010年)

摘要

民國95~96年間由於信用卡與現金卡呆帳金額嚴重惡化形成所謂之卡債風暴，為舒緩卡債對於社會之衝擊，主管機關與銀行間制定許多紓困之還款方案，本文針對銀行自行協商客戶進行違約因素分析，並以Logistic模型建構違約預測模型及相關模型穩健度及準確度檢定，最終亦將模型評分結果轉換為信用評分(Score Bank)，藉以導入信用卡債權管理實務運用中。模型風險變數篩選除了參考文獻中顯著影響違約的風險變數，並加入銀行內部資料以及外部的信用資訊作為建置模型的基礎，主要發現及結論說明如下:
一、影響信用卡違約的風險變數，經過驗證共有34個顯著影響債務協商之違約風險，分別為年紀介於41~60歲之間、居住區域位於北中南部、性別為男性、教育程度為高中、申辦債務協商時延滯階段為90天~150天、協商分期期數介於1~24期及49~84期、無協商利率、協商繳款金額小於1,000元、信用卡持卡時間介於1~4年、年收入介於50~200之間、持卡張數1~3張及大於10張以上、整體無擔保負債介於50~200萬之間、DTI 41~50倍、持有現金卡。
二、本研究為強化模型之可信度與穩健度，特別進行多項檢定，其中模型準確預測比率達64.84%，樣本外測試(Out of Sample Testing)準確預測能力僅略為下降0.84%，模型配適度之HL擬合優度指標與模型相關係數Wald檢定均符合標準，鑑別力指標K-S值亦有31%達到鑑別能力中等的標準。
三、Logistic Model所產出的相關係數，需進行發生機率轉換與聯立方程式求解而得出各變數之分佈與信用評分，本研究已初步將客戶區分為五個風險等級，針對不同的客戶區隔，可運用於信用卡債權維護管理。

關鍵字

信用卡違約風險模型 ； Logistic模型 ； 債務協商

並列摘要

This study is to analyze the factors contributing to the credit card default from the cases of the bank’s payment plan, to construct the model of default prediction, and to ensure the stability and accuracy of related models by using the Logistic Model. The results from the model evaluation will be transferred into Score Bank in the hope of introducing them into the practical application of credit card collection management. Hence, the foundations of the model development are based upon risk variables investigated by academic research as well as bank’s internal data and external credit information. The main findings and conclusions are described as follows:
First, there are thirty-four risk variables clearly proved to affect the default of debt counseling, such as the 41-60 age group of card holders, residence area in the north and midland and south of Taiwan, male card holders, senior high school as the educational levels of card holders, delinquent status in debt counseling from 90 to 150 days, plans of debt negotiation from 1 to 24, 49 to 84 installments, free interest rate for debt counseling, payments for debt negotiation less than one thousand NT dollars, the duration of holding a credit card measuring from one to four years, annual income ranging from NT500,000 to NT2,000,000, possessing several credit cards at a time or even more than 10, total unsecured loan amount ranging between NT500,000 and NT2,000,000, DTI 41~50 times, using cash cards.
Second, this study conducts several trials to enhance the credibility and stability of the Logistic Model. Logistic Model’s accuracy of prediction reaches 64.84% and slightly decreases in 0.84% for Out of Sample Testing. Hosmer- Lemeshow Goodness of fit index and Wald examination both live up to the standard. K-S value, the index of discrimination, achieves the middle level at 31%.
Third, the related factors that Logistic Model generate will practice the probability transform and enter the simultaneous equations to produce the distribution of variables and credit evaluation. This study has initially divided credit card customers into five risk variables to suit different customer segmentations and proved to be applicable to the credit card collection management.

並列關鍵字

The risk model of credit card default ； Logistic Model ； Debt Consultation

參考文獻

11.莊瑞珠，2007，「邏輯斯迴歸模型運用在女性信用卡評分制度之研究」，輔仁管理評論，第十四卷，第一期，頁127-154。

15.郭淑薇，2006，「現金卡信用評分模型之建立與風險管理」，國立中央大學財務金融研究所碩士論文。

6.Tomas, L. C. and David, B.E. and Crook, N. J., 2002, “Statistical Methods for Building Credit Scorecards ”,Credit Scoring and its Applications,41-60.

8.Steenackers, A. and Goovaerts, M.J., 1989, “A Credit Scoring Model for Personal Loan Insurance: Mathematics and Economic ”, Vol.8,NO.31, 12-21.

一、國內文獻